search

gliese581gg / YOLO_tensorflow

tensorflow implementation of 'YOLO : Real-Time Object Detection'
Other
1.72k stars 656 forks source link

Hi, would you like to tell me why no softmax function used ? #29

Closed zh0ngtian closed 7 years ago

zh0ngtian commented 7 years ago

Hi, would you like to tell me why no softmax function used ? And would you like to share your cfg file ?

shamanez commented 7 years ago

Read the paper. There is a different architecture for score function layer. It's like calculating a tensor. This code is for actually test run. Author has clearly mentioned it :)

zh0ngtian commented 7 years ago

Thank you for your answer !

shamanez commented 7 years ago

well it's like that. They needed to have a softmax function for class classification it will be costly. I had that problem too :D :D. It's not so clear. But if we find how they calculate P(Object) for each grid cell we might come up with an answer. I think those probabilities clearly align with IOU of bounding boxes.

On Wed, Apr 26, 2017 at 12:27 PM, YaoZhongtian notifications@github.com wrote:

Closed #29 https://github.com/gliese581gg/YOLO_tensorflow/issues/29.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/gliese581gg/YOLO_tensorflow/issues/29#event-1057956847, or mute the thread https://github.com/notifications/unsubscribe-auth/AQHCmsqKBkfzRG0ivlcguKVhG-fZoqBTks5rzurUgaJpZM4NE88L .